Searcharxiv⌕ Search

arXiv subjects

Marius Junge

Publications and source records attributed to Marius Junge.

122 records · Page 7Linked to original sources

The norm of sums of independent noncommutative random variables in $L_p(\ell_1)$

We investigate the norm of sums of independent vector-valued random variables in noncommutative Lp spaces. This allows us to obtain a uniform family of complete embeddings of the Schatten class Sq^n in Sp(lq^m) with optimal order m = n^2. Using these embeddings we show the surprising fact that the sharp type (cotype) index in the sense of operator spaces for Lp[0,1] is min(p,p') (max(p,p')). Similar techniques are used to show that the operator space notions of B-convexity and K-convexity are equivalent.

math.FA↗

A classification for 2-isometries of noncommutative Lp-spaces

In this paper we extend previous results of Banach, Lamperti and Yeadon on isometries of Lp-spaces to the non-tracial case first introduced by Haagerup. Specifically, we use operator space techniques and an extrapolation argument to prove that every 2-isometry T : Lp(M) to Lp(N) between arbitrary noncommutative Lp-spaces can always be written in the form T(phi^{1/p}) = w (phi circ pi^{-1} circ E)^{1/p}, for phi in M_*^+. Here pi is a normal *-isomorphism from M onto the von Neumann subalgebra pi(M) of N, w is a partial isometry in N, and E is a normal conditional expectation from N onto pi(M). As a consequence of this, any 2-isometry is automatically a complete isometry and has completely contractively complemented range.

math.OA↗

Embedding of the operator space OH and the logarithmic `little Grothendieck inequality'

Using free random varaibles we find an embedding of the operator space $OH$ in the predual of a von Neumann algebra. The properties of this embedding allow us to determined the projection constant of $OH_n$, i.e. there exists a projection $P:B(\ell_2)\to OH_n$ whose completely bounded norm behaves as n^{1/2}/(1+ln n)^{1/2}. According to recent results of Pisier/Shlyahtenko, the lower bound holds for every projection. Improving a previous estimate of order $(1+ ln n)$ of the author, Pisier/Shlyahtenko obtained a `logarithmic little Grothendieck inequality'. We find a second proof of this inequality which explains why the factor $\sqrt{1+\ln n}$ is indeed necessary. In particular the operator space version of the `little Grothendieck inequality' fails to hold. This `logarithmic little Grothendieck' inequality characterizes $C^*$-algebras with the weak expectation property of Lance.

math.OA↗

The optimal order for the p-th moment of sums of independent random variables with respect to symmetric norms and related combinatorial estimates

We calculate the p-the moment of the sum of n independent random variables with respect to symmetric norm in R^n. The order of growth for upper bound p/ln p obtained in ths estimate is optimal. The result extends to generalized Lorentz spaces l_{f,w} under mild assumptions on f. Indeed, the key combinatorial estimate is obtained for the weak l_1 (l_{1,infinity})-norm. Similar results have been obtained independently by Gordon, Litvak, Schuett and Werner for Orlicz norms and by Montgomery-Smith using different techniques and avoiding the combinatorial estimate.

math.PR↗

Integral mappings and the principle of local reflexivity for noncommutative L^1-spaces

The operator space analogue of the {\em strong form} of the principle of local reflexivity is shown to hold for any von Neumann algebra predual, and thus for any $C^{*}$-algebraic dual. This is in striking contrast to the situation for $C^{*}$-algebras, since, for example, $K(H)$ does not have that property. The proof uses the Kaplansky density theorem together with a careful analysis of two notions of integrality for mappings of operator spaces.

math.OA↗

Orlicz property of operator spaces and eigenvalue estimates

As is well known absolute convergence and unconditional convergence for series are equivalent only in finite dimensional Banach spaces. Replacing the classical notion of absolutely summing operators by the notion of 1 summing operators \[ \summ_k || Tx_k || \leq c || \summ_k e_k \otimes x_k ||_{\ell_1\otimes_{min}E}\] in the category of operator spaces, it turns out that there are quite different interesting examples of 1 summing operator spaces. Moreover, the eigenvalues of a composition $TS$ decreases of order $n^{\frac{1}{q}}$ for all operators $S$ factorizing completely through a commutative $C^*$-algebra if and only if the 1 summing norm of the operator $T$ restricted to a $n$-dimensional subspace is not larger than $c n^{1-\frac{1}{q}}$, provided $q>2$. This notion of 1 summing operators is closely connected to the notion of minimal and maximal operator spaces.

math.FA↗

Type and cotype with respect to arbitrary orthonormal systems

Let $\on_{k \in \nz}$ be an orthonormal system on some $σ$-finite measure space $(\Om,p)$. We study the notion of cotype with respect to $Φ$ for an operator $T$ between two Banach spaces $X$ and $Y$, defined by $\fco T := \inf$ $c$ such that \[ \Tfmm \pl \le \pl c \pll \gmm \hspace{.7cm}\mbox{for all}\hspace{.7cm} (x_k)\subset X \pl,\] where $(g_k)_{k\in \nz}$ is a sequence of independent and normalized gaussian variables. It is shown that this $Φ$-cotype coincides with the usual notion of cotype $2$ iff \linebreak $\fco {I_{\lin}} \sim \sqrt{\frac{n}{\log (n+1)}}$ uniformly in $n$ iff there is a positive $η>0$ such that for all $n \in \nz$ one can find an orthonormal $Ψ= (ψ_l)_1^n \subset {\rm span}\{ ϕ_k \p|\p k \in \nz\}$ and a sequence of disjoint measurable sets $(A_l)_1^n \subset \Om$ with \[ \int\limits_{A_l} \bet ψ_l\rag^2 d p \pl \ge \pl η\quad \mbox{for all}\quad l=1,...,n \pl. \] A similar result holds for the type situation. The study of type and cotype with respect to orthonormal systems of a given length provides the appropriate approach to this result. We intend to give a quite complete picture for orthonormal systems in measure space with few atoms.

math.FA↗

Proportional subspaces of spaces with unconditional basis have good volume properties

A generalization of Lozanovskii's result is proved. Let E be $k$-dimensional subspace of an $n$-dimensional Banach space with unconditional basis. Then there exist $x_1,..,x_k \subset E$ such that $B_E \p \subset \p absconv\{x_1,..,x_k\}$ and \[ \kla \frac{{\rm vol}(absconv\{x_1,..,x_k\})}{{\rm vol}(B_E)} \mer^{\frac{1}{k}} \kl \kla e\p \frac{n}{k} \mer^2 \pl .\] This answers a question of V. Milman which appeared during a GAFA seminar talk about the hyperplane problem. We add logarithmical estimates concerning the hyperplane conjecture for proportional subspaces and quotients of Banach spaces with unconditional basis. File Length:27K

math.FA↗

Cotype and summing properties in Banach spaces

It is well known in Banach space theory that for a finite dimensional space $E$ there exists a constant $c_E$, such that for all sequences $(x_k)_k \subset E$ one has \[ \summ_k \noo x_k \rrm \kl c_E \pl \sup_{\eps_k \pm 1} \noo \summ_k \eps_k x_k \rrm \pl .\] Moreover, if $E$ is of dimension $n$ the constant $c_E$ ranges between $\sqrt{n}$ and $n$. This implies that absolute convergence and unconditional convergence only coincide in finite dimensional spaces. We will characterize Banach spaces $X$, where the constant $c_E \sim \sqrt{n}$ for all finite dimensional subspaces. More generally, we prove that an estimate $c_E \kll c n^{1-\frac{1}{q}}$holds for all $n \in \nz$ and all $n$-dimensional subspaces $E$ of $X$ if and only if the eigenvalues of every operator factoring through $\ell_{\infty}$ decrease of order $k^{-\frac{1}{q}}$ if and only if $X$ is of weak cotype $q$, introduced by Pisier and Mascioni. We emphasize that in contrast to Talagrand's equivalence theorem on cotype $q$ and absolutely $(q,1)$-summing spaces this extendsto the case $q=2$. If $q>2$ and one of the conditions above is satisfied one has \[ \kla \summ_k \noo x_k \rrm^q \mer^{\frac{1}{q}} \kl C^{1+l}\pl (1+{\rm log}_2)^{(l)}((1 +{\rm log}_2 n)^{\frac{1}{q}}) \pl \ez \noo \summ_k \eps_k x_k \rrm \] for all $n,l \in \nz$ and $(x_k)_k \subset E$, $E$ a $n$ dimensional subspace of $X$. In the case $q=2$ the same holds if we replace the expected value by the supremum.

math.FA↗

Hyperplane conjecture for quotient spaces of $L_p$

We give a positive solution for the hyperplane conjecture of quotient spaces F of $L_p$, where $1<p\kll\infty$. \[ vol(B_F)^{\frac{n-1}{n}} \kl c_0 \pl p' \pl \sup_{H \p hyperplane} vol(B_F\cap H) \pl.\] This result is extended to Banach lattices which does not contain $\ell_1^n$'s uniformly. Our main tools are tensor products and minimal volume ratio with respect to $L_p$-sections.

math.FA↗

Bilinear forms on exact operator spaces and B(H)\otimes B(H)

Let $E,F$ be exact operators (For example subspaces of the $C^*$-algebra $K(H)$ of all the compact operators on an infinite dimensional Hilbert space $H$). We study a class of bounded linear maps $u\colon E\to F^*$ which we call tracially bounded. In particular, we prove that every completely bounded (in short $c.b.$) map $u\colon E\to F^*$ factors boundedly through a Hilbert space. This is used to show that the set $OS_n$ of all $n$-dimensional operator spaces equipped with the $c.b.$ version of the Banach Mazur distance is not separable if $n>2$. As an application we show that there is more than one $C^*$-norm on $B(H)\otimes B(H)$, or equivalently that $$B(H)\otimes_{\min}B(H)\not=B(H)\otimes_{\max}B(H),$$ which answers a long standing open question. Finally we show that every ``maximal" operator space (in the sense of Paulsen) is not exact in the infinite dimensional case, and in the finite dimensional case, we give a lower bound for the ``exactness constant".

math.FA↗

Comparing gaussian and Rademacher cotype for operators on the space of continous functions

We will prove an abstract comparision principle which translates gaussian cotype in Rademacher cotype conditions and vice versa. More precisely, let $2\!<\!q\!<\!\infty$ and $T:\,C(K)\,\to\,F$ a linear, continous operator. T is of gaussian cotype q if and only if ( \summ_1^n (\frac{|| Tx_k||_F}{\sqrt{\log(k+1)}})^q )^{1/q} \, \le c || \summ_1^n \varepsilon_k x_k ||_{L_2(C(K))} , for all sequences with $(|| Tx_k ||)_1^n$ decreasing. T is of Rademacher cotype q if and only if (\summ_1^n (|| Tx_k||_F \,\sqrt{\log(k+1)})^q )^{1/q} \, \le c || \summ_1^n g_k x_k ||_{L_2(C(K))} , for all sequences with $(||Tx_k ||)_1^n$ decreasing. Our methods allows a restriction to a fixed number of vectors and complements the corresponding results of Talagrand.

math.FA↗